A64: Add Step

Allow for stepping instruction-by-instruction

include/dynarmic/A64/a64.h

@@ -30,6 +30,12 @@ public:
      */
     void Run();
 
+    /**
+     * Step the emulated CPU for one instruction.
+     * Cannot be recursively called.
+     */
+    void Step();
+
     /**
      * Clears the code cache of all compiled code.
      * Can be called at any time. Halts execution if called within a callback.

src/backend/x64/a64_emit_x64.cpp

@@ -291,11 +291,11 @@ void A64EmitX64::GenTerminalHandlers() {
         // This calculation has to match up with A64::LocationDescriptor::UniqueHash
         // TODO: Optimization is available here based on known state of fpcr.
         code.mov(rbp, qword[r15 + offsetof(A64JitState, pc)]);
-        code.mov(rcx, A64::LocationDescriptor::PC_MASK);
+        code.mov(rcx, A64::LocationDescriptor::pc_mask);
         code.and_(rcx, rbp);
         code.mov(ebx, dword[r15 + offsetof(A64JitState, fpcr)]);
-        code.and_(ebx, A64::LocationDescriptor::FPCR_MASK);
+        code.and_(ebx, A64::LocationDescriptor::fpcr_mask);
-        code.shl(ebx, 37);
+        code.shl(ebx, A64::LocationDescriptor::fpcr_shift);
         code.or_(rbx, rcx);
     };
 

src/backend/x64/a64_interface.cpp

@@ -54,13 +54,28 @@ public:
 
         // TODO: Check code alignment
 
-        const u32 new_rsb_ptr = (jit_state.rsb_ptr - 1) & A64JitState::RSBPtrMask;
+        const CodePtr current_code_ptr = [this]{
-        if (jit_state.GetUniqueHash() == jit_state.rsb_location_descriptors[new_rsb_ptr]) {
+            // RSB optimization
-            jit_state.rsb_ptr = new_rsb_ptr;
+            const u32 new_rsb_ptr = (jit_state.rsb_ptr - 1) & A64JitState::RSBPtrMask;
-            block_of_code.RunCodeFrom(&jit_state, reinterpret_cast<CodePtr>(jit_state.rsb_codeptrs[new_rsb_ptr]));
+            if (jit_state.GetUniqueHash() == jit_state.rsb_location_descriptors[new_rsb_ptr]) {
-        } else {
+                jit_state.rsb_ptr = new_rsb_ptr;
-            block_of_code.RunCode(&jit_state);
+                return reinterpret_cast<CodePtr>(jit_state.rsb_codeptrs[new_rsb_ptr]);
-        }
+            }
+
+            return GetCurrentBlock();
+        }();
+        block_of_code.RunCodeFrom(&jit_state, current_code_ptr);
+
+        PerformRequestedCacheInvalidation();
+    }
+
+    void Step() {
+        ASSERT(!is_executing);
+        is_executing = true;
+        SCOPE_EXIT { this->is_executing = false; };
+        jit_state.halt_requested = true;
+
+        block_of_code.StepCode(&jit_state, GetCurrentSingleStep());
 
         PerformRequestedCacheInvalidation();
     }
@@ -185,9 +200,19 @@ private:
         return this_->GetCurrentBlock();
     }
 
-    CodePtr GetCurrentBlock() {
+    IR::LocationDescriptor GetCurrentLocation() const {
-        IR::LocationDescriptor current_location{jit_state.GetUniqueHash()};
+        return IR::LocationDescriptor{jit_state.GetUniqueHash()};
+    }
 
+    CodePtr GetCurrentBlock() {
+        return GetBlock(GetCurrentLocation());
+    }
+
+    CodePtr GetCurrentSingleStep() {
+        return GetBlock(A64::LocationDescriptor{GetCurrentLocation()}.SetSingleStepping(true));
+    }
+
+    CodePtr GetBlock(IR::LocationDescriptor current_location) {
         if (auto block = emitter.GetBasicBlock(current_location))
             return block->entrypoint;
 
@@ -256,6 +281,10 @@ void Jit::Run() {
     impl->Run();
 }
 
+void Jit::Step() {
+    impl->Step();
+}
+
 void Jit::ClearCache() {
     impl->ClearCache();
 }

src/backend/x64/a64_jitstate.cpp

@@ -10,12 +10,6 @@
 
 namespace Dynarmic::BackendX64 {
 
-u64 A64JitState::GetUniqueHash() const noexcept {
-    const u64 fpcr_u64 = static_cast<u64>(fpcr & A64::LocationDescriptor::FPCR_MASK) << 37;
-    const u64 pc_u64 = pc & A64::LocationDescriptor::PC_MASK;
-    return pc_u64 | fpcr_u64;
-}
-
 /**
  * Comparing MXCSR and FPCR
  * ========================

src/backend/x64/a64_jitstate.h

@@ -11,6 +11,7 @@
 #include <xbyak.h>
 
 #include "common/common_types.h"
+#include "frontend/A64/location_descriptor.h"
 
 namespace Dynarmic::BackendX64 {
 
@@ -79,7 +80,11 @@ struct A64JitState {
     void SetFpcr(u32 value);
     void SetFpsr(u32 value);
 
-    u64 GetUniqueHash() const noexcept;
+    u64 GetUniqueHash() const noexcept {
+        const u64 fpcr_u64 = static_cast<u64>(fpcr & A64::LocationDescriptor::fpcr_mask) << A64::LocationDescriptor::fpcr_shift;
+        const u64 pc_u64 = pc & A64::LocationDescriptor::pc_mask;
+        return pc_u64 | fpcr_u64;
+    }
 };
 
 #ifdef _MSC_VER

src/backend/x64/block_of_code.cpp

@@ -142,6 +142,10 @@ void BlockOfCode::RunCodeFrom(void* jit_state, CodePtr code_ptr) const {
     run_code_from(jit_state, code_ptr);
 }
 
+void BlockOfCode::StepCode(void* jit_state, CodePtr code_ptr) const {
+    step_code(jit_state, code_ptr);
+}
+
 void BlockOfCode::ReturnFromRunCode(bool mxcsr_already_exited) {
     size_t index = 0;
     if (mxcsr_already_exited)
@@ -174,6 +178,19 @@ void BlockOfCode::GenRunCode() {
     SwitchMxcsrOnEntry();
     jmp(r14);
 
+    align();
+    step_code = getCurr<RunCodeFromFuncType>();
+
+    ABI_PushCalleeSaveRegistersAndAdjustStack(*this);
+
+    mov(r15, ABI_PARAM1);
+
+    mov(qword[r15 + jsi.offsetof_cycles_to_run], 1);
+    mov(qword[r15 + jsi.offsetof_cycles_remaining], 1);
+
+    SwitchMxcsrOnEntry();
+    jmp(ABI_PARAM2);
+
     align();
     run_code = getCurr<RunCodeFuncType>();
 

src/backend/x64/block_of_code.h

@@ -51,6 +51,8 @@ public:
     void RunCode(void* jit_state) const;
     /// Runs emulated code from code_ptr.
     void RunCodeFrom(void* jit_state, CodePtr code_ptr) const;
+    /// Runs emulated code from code_ptr for a single cycle.
+    void StepCode(void* jit_state, CodePtr code_ptr) const;
     /// Code emitter: Returns to dispatcher
     void ReturnFromRunCode(bool mxcsr_already_exited = false);
     /// Code emitter: Returns to dispatcher, forces return to host
@@ -158,6 +160,7 @@ private:
     using RunCodeFuncType = void(*)(void*);
     using RunCodeFromFuncType = void(*)(void*, CodePtr);
     RunCodeFuncType run_code = nullptr;
+    RunCodeFromFuncType step_code = nullptr;
     RunCodeFromFuncType run_code_from = nullptr;
     static constexpr size_t MXCSR_ALREADY_EXITED = 1 << 0;
     static constexpr size_t FORCE_RETURN = 1 << 1;

src/common/bit_util.h

@@ -23,7 +23,7 @@ constexpr size_t BitSize() {
 }
 
 template <typename T>
-inline T Ones(size_t count) {
+constexpr T Ones(size_t count) {
     ASSERT_MSG(count <= BitSize<T>(), "count larger than bitsize of T");
     if (count == BitSize<T>())
         return static_cast<T>(~static_cast<T>(0));

src/frontend/A64/location_descriptor.cpp

@@ -12,7 +12,7 @@
 namespace Dynarmic::A64 {
 
 std::ostream& operator<<(std::ostream& o, const LocationDescriptor& descriptor) {
-    o << fmt::format("{{{}, {}}}", descriptor.PC(), descriptor.FPCR().Value());
+    o << fmt::format("{{{}, {}{}}}", descriptor.PC(), descriptor.FPCR().Value(), descriptor.SingleStepping() ? ", step" : "");
     return o;
 }
 

src/frontend/A64/location_descriptor.h

@@ -24,19 +24,29 @@ namespace Dynarmic::A64 {
  */
 class LocationDescriptor {
 public:
-    static constexpr u64 PC_MASK = 0x00FF'FFFF'FFFF'FFFFull;
+    static constexpr size_t pc_bit_count = 56;
-    static constexpr u32 FPCR_MASK = 0x07C8'0000;
+    static constexpr u64 pc_mask = Common::Ones<u64>(pc_bit_count);
+    static constexpr u32 fpcr_mask = 0x07C8'0000;
+    static constexpr size_t fpcr_shift = 37;
+    static constexpr size_t single_stepping_bit = 57;
+    static_assert((pc_mask & (u64(fpcr_mask) << fpcr_shift) & (u64(1) << single_stepping_bit)) == 0);
 
-    LocationDescriptor(u64 pc, FP::FPCR fpcr) : pc(pc & PC_MASK), fpcr(fpcr.Value() & FPCR_MASK) {}
+    LocationDescriptor(u64 pc, FP::FPCR fpcr, bool single_stepping = false)
+        : pc(pc & pc_mask), fpcr(fpcr.Value() & fpcr_mask), single_stepping(single_stepping)
+    {}
 
     explicit LocationDescriptor(const IR::LocationDescriptor& o)
-        : pc(o.Value() & PC_MASK), fpcr((o.Value() >> 37) & FPCR_MASK) {}
+        : pc(o.Value() & pc_mask)
+        , fpcr((o.Value() >> fpcr_shift) & fpcr_mask)
+        , single_stepping(Common::Bit<single_stepping_bit>(o.Value()))
+    {}
 
-    u64 PC() const { return Common::SignExtend<56>(pc); }
+    u64 PC() const { return Common::SignExtend<pc_bit_count>(pc); }
     FP::FPCR FPCR() const { return fpcr; }
+    bool SingleStepping() const { return single_stepping; }
 
     bool operator == (const LocationDescriptor& o) const {
-        return std::tie(pc, fpcr) == std::tie(o.pc, o.fpcr);
+        return std::tie(pc, fpcr, single_stepping) == std::tie(o.pc, o.fpcr, single_stepping);
     }
 
     bool operator != (const LocationDescriptor& o) const {
@@ -44,18 +54,23 @@ public:
     }
 
     LocationDescriptor SetPC(u64 new_pc) const {
-        return LocationDescriptor(new_pc, fpcr);
+        return LocationDescriptor(new_pc, fpcr, single_stepping);
     }
 
     LocationDescriptor AdvancePC(int amount) const {
-        return LocationDescriptor(static_cast<u64>(pc + amount), fpcr);
+        return LocationDescriptor(static_cast<u64>(pc + amount), fpcr, single_stepping);
+    }
+
+    LocationDescriptor SetSingleStepping(bool new_single_stepping) const {
+        return LocationDescriptor(pc, fpcr, new_single_stepping);
     }
 
     u64 UniqueHash() const noexcept {
         // This value MUST BE UNIQUE.
         // This calculation has to match up with EmitTerminalPopRSBHint
-        const u64 fpcr_u64 = static_cast<u64>(fpcr.Value()) << 37;
+        const u64 fpcr_u64 = static_cast<u64>(fpcr.Value()) << fpcr_shift;
-        return pc | fpcr_u64;
+        const u64 single_stepping_u64 = static_cast<u64>(single_stepping) << single_stepping_bit;
+        return pc | fpcr_u64 | single_stepping_u64;
     }
 
     operator IR::LocationDescriptor() const {
@@ -65,6 +80,7 @@ public:
 private:
     u64 pc;        ///< Current program counter value.
     FP::FPCR fpcr; ///< Floating point control register.
+    bool single_stepping;
 };
 
 /**

src/frontend/A64/translate/translate.cpp

@@ -9,6 +9,7 @@
 #include "frontend/A64/translate/impl/impl.h"
 #include "frontend/A64/translate/translate.h"
 #include "frontend/ir/basic_block.h"
+#include "frontend/ir/terminal.h"
 
 namespace Dynarmic::A64 {
 
@@ -16,8 +17,9 @@ IR::Block Translate(LocationDescriptor descriptor, MemoryReadCodeFuncType memory
     IR::Block block{descriptor};
     TranslatorVisitor visitor{block, descriptor, std::move(options)};
 
+    const bool single_step = descriptor.SingleStepping();
     bool should_continue = true;
-    while (should_continue) {
+    do {
         const u64 pc = visitor.ir.current_location->PC();
         const u32 instruction = memory_read_code(pc);
 
@@ -29,6 +31,10 @@ IR::Block Translate(LocationDescriptor descriptor, MemoryReadCodeFuncType memory
 
         visitor.ir.current_location = visitor.ir.current_location->AdvancePC(4);
         block.CycleCount()++;
+    } while (should_continue && !single_step);
+
+    if (single_step && should_continue) {
+        visitor.ir.SetTerm(IR::Term::LinkBlock{*visitor.ir.current_location});
     }
 
     ASSERT_MSG(block.HasTerminal(), "Terminal has not been set");